Some time ago I attempted to use VirtualDub's temporal filter to clean up some VHS caps, but I was getting an awful residual image after every scene change - similar to what I noticed in the Cowclops V2 DVD screenshots, but worse.
Is the filter you used the same as the AVISynth TemporalSmoother function, which is apparently a direct port from VDub? If so, then you could probably refine your process to a single AVISynth script, saving on the amount of hard drive space required.
Here's a relevant thread on Doom9, which explains the difference betweeen "bidirectional" filtering used by TemporalSoften, for example, and a true forwards/backwards average function like you have written.
It'd be interesting to see how your method compares against other popular filters.