We've recently gotten to do (and see results of) some testing around internal links & anchor text and have come to some interesting, if not 100% proven, conclusions. As a disclaimer, SEO testing, whether done in a controlled environment or on live sites, is both challenging to quantify and subject to fair critiques both practically and academically. That said, I feel fairly confident in stating the following findings.

NOTE: These hold true primarily (and in some cases exclusively) for Google. I don't have enough data/experience to say whether the same results would be found at Yahoo! or MSN/Live.

1. Internal Anchor Text Has Very Little Impact on the Homepage
   Although linking to a page like our Web 2.0 Awards with the anchor text "web 2.0," for example, appears to provide a positive nudge, the same does not hold true if I link to the homepage of SEOmoz with the anchor text "SEO." In fact, for those who put a great deal of time into optimizing their links back to their homepage to say "new york apartments" or "antique rifles" rather than "home," I'd test whether modification of the link anchor text has any adverse rankings impact. In my recent experience (on a few different sites), the answer is no.
   _
2. Excessive Internal Anchor Text Linking / Manipulation Can Trip An Automated Penalty on Google
   I recently had my second run-in with a penalty at Google that appears to punish sites for excessive internal linking with "optimized" (or "keyword stuffed anchor text") links. When the links were removed (in both cases, they were found in the footer of the website sitewide), the rankings were restored immediately following Google's next crawl, indicating a fully automated filter (rather than a manual penalty requiring a re-consideration request).
   _
3. Beyond a Certain Point, Adding More Internal Links to a Page Does Not Necessarily Flow More Anchor Text Value
   Adding internal links from many important pages does provide ranking benefit, but it appears not to matter whether those links contain optimized anchor text after the first few (or few dozen, depending on site size). Perhaps there's a limit to how much anchor text value can flow from a single site to itself (or to another site, which might explain the diminished value of sitewide links) or maybe Google discounts optimized internal anchor text after a certain point. Whatever the reason, my general suggestion going forward is to link to pages with optimized links to help with link juice flow, but not necessarily obsess over anchor text (at least, until I see data that says otherwise).
   
   BTW - This doesn't mean I'm abandoning my recommendations for internal linking with good anchor text for either usability, keyword cannibalization or to provide an initial boost. It just means sitewides (or pouring links from 500 pages instead of 50) aren't going to make the "recommended list" of SEO tactics for many future clients.

I'd love to hear your experiences in both testing and live scenarios and any speculation you've got as to why these phenomena appear to hold true. As always, I recommend conducting your own tests and applying what shows the best results for your sites.

BTW - For those interested in SEO experimentation, I've described the process in the past here and here.