But anyway, what nimrody suggested is what they are doing - using platform neutral Dalvik bytecode and then transition to native through JIT.
Really, the existing non-jit'ed Dalvik VM runs fast enough. If you think about it, their decision to ship a non-JIT VM and enhance the platform with JIT later is very much like Apple's approach with the iPhone of initially shipping the product without cut-and-paste and without any native API for third party apps, but then enhancing the platform later by adding those (and many other) features later. Both companies shipped when they had a platform that was "good enough", then have kept iterating and improving.
I'm looking forward to seeing Bill Buzbee's talk on the JIT at I/O (of the most excellent http://www.homebrewcpu.com/ incidentally! :)
The Dalvik VM uses its own bytecode and all, so they would have to heavily modify anything existing.
Ultimately Android has been a work in progress. JIT is something that everyone has been asking for since day one, but there just aren't enough resources -- Apple is a $220 billion dollar company that is significantly hinged on the iPhone OS. The Android project, while relied upon by a lot of companies, is largely the work of Google that sees little to no revenue from it (their primary motivation, purportedly, was simply to ensure that there were options in the market).
That's actually an interesting facet of Android -- much was made recently about one of the open handset partners replacing Google with another search engine, but that's exactly how the platform is supposed to work. Google has no free in to all of the platform, beyond the add-ons like Google Mail that the partners would almost certainly want on the phone.