multi-threaded encoder generates different output than single-threaded

[shekh]

I noticed some pattern, therefore looks suspicious.
Encoding constant input (repeated still frame)
single-threaded encoder output:
A,B,C,C,C,...
multi-threaded (2 threads):
A,A,B,B,C,C,C,...
multi-threaded (4 threads):
A,A,A,A,B,B,B,B,C,C,C,...
So is the output affected by some internal state in CSampleEncoder? Should this happen? Should this state be shared in encoders pool?
Decoded differences are somewhere 1-bit high.

[dnewman-gpsw]

Only the metadata should be slightly different per frame (timecode and frame number -- both used for Active Metadata.) Nothing else should be changing. Could be a bug.