diff --git a/ChangeLog b/ChangeLog
index a9db11d..b93a2e5 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -3,6 +3,7 @@ Unreleased
 	* SemiBin1: Introduce separate SemiBin1 command
 	* internal: Code simplification and refactor
 	* deprecation: Deprecate --orf-finder=fraggenescan option
+	* SemiBin: do not use more processes than can be taken advantage of (#155)
 
 Version 2.0.2 Oct 31 2023 by BigDataBiology
 	* multi_easy_bin: Fix multi_easy_bin with --write-pre-recluster (#128)
diff --git a/SemiBin/cluster.py b/SemiBin/cluster.py
index 791fca2..dc9759a 100644
--- a/SemiBin/cluster.py
+++ b/SemiBin/cluster.py
@@ -17,8 +17,8 @@ def run_infomap(g, edge_weights, vertex_weights, num_process):
     '''Run infomap, using multiple processors (if available)'''
     if num_process == 1:
         return g.community_infomap(edge_weights=edge_weights, vertex_weights=vertex_weights, trials=NR_INFOMAP_TRIALS)
-    import multiprocessing
-    with multiprocessing.Pool(num_process) as p:
+    import multiprocessing as mp
+    with mp.Pool(min(num_process, NR_INFOMAP_TRIALS)) as p:
         rs = [p.apply_async(run_infomap1, (g, edge_weights, vertex_weights, 1))
                 for _ in range(NR_INFOMAP_TRIALS)]
         p.close()
diff --git a/SemiBin/main.py b/SemiBin/main.py
index e35d808..125fca6 100644
--- a/SemiBin/main.py
+++ b/SemiBin/main.py
@@ -773,6 +773,44 @@ def generate_sequence_features_single(logger, contig_fasta,
         logger.info('We will only calculate k-mer features.')
 
     if not only_kmer:
+        n_sample = len(bams)
+        is_combined = n_sample >= 5
+        bam_list = bams
+
+        logger.info('Calculating coverage for every sample.')
+
+        with Pool(min(num_process, len(bams))) as pool:
+            results = [
+                pool.apply_async(
+                    generate_cov,
+                    args=(
+                        bam_file,
+                        bam_index,
+                        output,
+                        must_link_threshold,
+                        is_combined,
+                        binned_length,
+                        logger,
+                        None
+                    ))
+                for bam_index, bam_file in enumerate(bams)]
+            for r in results:
+                s = r.get()
+                logger.info(f'Processed: {s}')
+
+        for bam_index, bam_file in enumerate(bams):
+            if not os.path.exists(os.path.join(output, '{}_data_cov.csv'.format(
+                    os.path.split(bam_file)[-1] + '_{}'.format(bam_index)))):
+                sys.stderr.write(
+                    f"Error: Generating coverage file fail\n")
+                sys.exit(1)
+            if is_combined:
+                if not os.path.exists(os.path.join(output, '{}_data_split_cov.csv'.format(
+                        os.path.split(bam_file)[-1] + '_{}'.format(bam_index)))):
+                    sys.stderr.write(
+                        f"Error: Generating coverage file fail\n")
+                    sys.exit(1)
+
         logger.debug('Start generating kmer features from fasta file.')
         kmer_whole = generate_kmer_features_from_fasta(
             contig_fasta, binned_length, 4)
@@ -922,6 +960,7 @@ def fasta_sample_iter(fn):
                                         os.path.join(args.output, f'samples/{sample}.fa'),
                                         args.ratio)
 
+
     if args.bams:
         with Pool(args.num_process if args.num_process != 0 else None) as pool:
             results = [
diff --git a/SemiBin/naive_orffinder.py b/SemiBin/naive_orffinder.py
index 2227ead..0f5791b 100644
--- a/SemiBin/naive_orffinder.py
+++ b/SemiBin/naive_orffinder.py
@@ -9,6 +9,12 @@
 STOP_CODONS = ['TAA', 'TAG', 'TGA']
 MIN_LEN = 90
 
+# See https://github.com/BigDataBiology/SemiBin/issues/155
+# A bit of testing showed that scaling tops up at 10~12 processes (on a 16-core
+# machine). This leaves a little leeway for a few more processes, but still
+# caps it to avoid #155
+MAX_PROCESS_ORFS = 24
+
 def reverse_complement(seq):
     return seq[::-1].translate(str.maketrans('ATGC', 'TACG'))
 
@@ -123,7 +129,7 @@ def run_naiveorf(fasta_path, num_process, output):
                     out.write(f'{translate(extract(seq, orf))}\n')
         else:
             ctx = mp.get_context('spawn')
-            with ctx.Pool(processes=num_process) as p:
+            with ctx.Pool(processes=min(num_process, MAX_PROCESS_ORFS)) as p:
                 outs = p.imap(get_orfs,
                         fasta.fasta_iter(fasta_path),
                         chunksize=8)
diff --git a/test/test_bin.py b/test/test_bin.py
index 900adee..f288f04 100644
--- a/test/test_bin.py
+++ b/test/test_bin.py
@@ -94,6 +94,14 @@ def test_cluster():
             res,
             [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
 
+def _renumber_cluster(xs):
+    ys = np.zeros_like(xs)
+    renumber = {}
+    for ix,x in enumerate(xs):
+        if x not in renumber:
+            renumber[x] = len(renumber)
+        ys[ix] = renumber[x]
+    return ys
 
 def test_recluster():
     contig_dict = {h:seq for h,seq in fasta_iter('test/bin_data/input.fasta')}
@@ -116,10 +124,12 @@ def test_recluster():
             random_seed=123)
 
     # Computed with a previous version
-    np.testing.assert_array_equal(
-            reclustered,
-            np.array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 29, 21, 22, 23, 24, 25, 26,
+    expected = np.array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 29, 21, 22, 23, 24, 25, 26,
                26, 28, 27, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 10, 11, 12, 13,
-               14, 15, 16, 17, 18, 19]))
+               14, 15, 16, 17, 18, 19])
+
+    np.testing.assert_array_equal(
+            _renumber_cluster(reclustered),
+            _renumber_cluster(expected))