:) I thought I recognised the name, and as a microbial genomics guy myself, I'm pretty familiar with TIGR. I suspect we agree more than disagree on OA, but these are some good points for debate. Genomics has largely been very pro-OA thanks to the efforts of the Wellcome Trust.
1. So I found this quite interesting. I think in the UK it's rare for an institution of that size to not be affiliated with a university and so able to get access. I was at the Sanger which was Cambridge Uni affiliated. However, this is also an issue of journal bundling and secret negotiations. Unless you're suggesting the complete dismantlement of the publishing scene, then this needs a complex solution. And re startups, I'm not sure why giving commercial entities subsidised access to the literature is a priority?
2) Full text analysis is not really an OA issue either (I've also not been very impressed with the results from it so far, but that's a different story). Yes it can be done more easily via OA, but most major publishers are increasingly supporting it. They can be obliged to support it without forcing OA as a publishing model. Many other issues fall under local copyright laws which makes it more complex. However, educational purposes and informal sharing is normally covered by fair use. And even with online textbooks, I'm not sure even if I publish as OA that I necessarily want someone else using it for commercial purposes for free. And licensing commercial use is a valid method for OA publishers to make money, surely (something they are struggling to do)? Of course, there are many different OA models with very varied licensing.
3) So I absolutely agree with improving public engagement, but increasingly methods sections are small print, very small word limit and stuck at the end of the paper. This utterly and fundamentally breaks the most important part of the scientific literature - reproducibility. And that is at crisis level now (ok hyperbole, but have you ever replicated something and not discovered that half the essential information is missing?). Similarly poor but simple/comprehensible papers will be higher rated than superior but complex papers.
We are asking the core literature to server two masters with very different needs, and my instinct is that this is ultimately not tenable. Review papers are far more useful to the public in general, since unless you're getting your hands dirty with the data, and learning it's foibles, then it's difficult to evaluate focussed research. This is especially true when dealing with sequence databases (assumptions of sequence accuracy, over/under representation etc), as I'm sure you appreciate.