cvs accidentaly breaks rcs file

cvs accidentaly breaks rcs file

I have an cvs server with pretty big binray file (about 50Mb) and its rcs
file is about 2G
Today I discovered that file update returns an error:
cvs [update aborted]: unexpected EOF reading /repository/cvsroot/dir/file,v
I checked the cvs server and it showed me an completely destroyed
/repository/cvsroot/dir/file,v with size about 100K.
I restored the file from backup, unfortunately I caught the same issue one
more time.
After second file replace now it acts fine, but I couldn't get the reason
of such behaviour.
It doesn't seem an hardware fail as all other files on the same disk are
fine and smartctl doesn't show any error or suspicious warnings.
Is there a way to catch the root of this issue?
ps. cvs is running on red hat 5.5, cvs version is 1.12.6.