Here's an edited version of a response I sent to him (short version: I like it !)
I think the time is ripe for this: it seems that people are getting more and more used to using the arxiv/iacr/eccc for tech reports and DBLP as a de facto list of papers, and even regularly subscribing to arxiv rss feeds to see what's new. bibref management systems like Mendeley/citeulike would also really benefit from this.
While (like others) I'm concerned about facilitating ranking schemes too much (I personally think the h-index is an abomination, but that's a different discussion), I think that even if the only outcome of this was to have a centralized single repository for CS publications, that in itself would be a major benefit.
I'm less sure about attention/reputation mechanisms though. It's clear that one of the challenges for researchers today is the 'eyeballs problem': how to get attention to your work amidst the sea of publications. While one might argue that Google and page-rank have done a good job of this, i think that over time it's become more and more top heavy, with a few locations acquiring sticky reputation and sucking in attention, and while this might be ok for general news, it's not so for research, where more often than not, good ideas can come from less "well known" sources.
I don't think CSPub causes any additional problems in this regard - but it would seem like much more thought is needed to design *transparent* ranking schemes. While google can do what they want with their ranking scheme, and keep it as a trade secret, a public service such as CSPub should try to keep ranking methods as transparent as possible. (hack-proof ranking methods ? I know there's research on this !)