Latest versions of Clamwin seem to have some sort of a leak/bug of not releasing win32 handles.
Scheduled scan is running for a few hours right now and TaskInfo shows clamwin currently using around 630,000 handles and the number is growing. All handles look like this: Type:Mutant, Refs:3/1, Name: <blank> (no name)

original clamav core has many mutex (then handles on win32) leaks since never calls destroy mutex, I really don't want to hunt all the concurrent clamav code to fix each of them I hope this is not a big deal

Weeeeeelll.... It didn't cause any real problems. As of yet. But such enourmous numbers of handles used by one program could potentially lead to some. It's definitely not a critical thing, but I'd say it would have to be addressed at some point. Maybe slowly one by one for those parts of code you're currently working on.

I've modified many parts of clamav code to make it a native port, but dealing with mutex is not trivial, so I need some time to look at, but since I'm not a developer of the original clamav I'm not happy to do such fixes
Anyway we are working on the next version of clamav that currently uses libclamav port, but we would to make a libclamav from scratch, obiviously not at the top of the todo list since there are a lot of things to do.
The current mutex/concurrent code works except for the handle leak, so I wouldn't anyway break the code.
I'm sure that is not really a file handle leak since an open handle will lead to spurious temp files in %temp% directory, and as far I tested clamscan leaves temp files only if unterrupted by user.