You should be able to go back and forth between "Enhanced" and "Maximized" without any reinstallation, but until you feel confident that no glitches or cache-corruption occurs over some reasonable period of time, you're advised to use "Enhanced."
The worst scenario with Maximized mode: you may have been in the middle of a large software installation, or for that matter -- any software installation.
Further, ISRT has a few glitches of its own, and Intel has released some two or three versions of IRST -- the software, Intel Rapid Storage Technology.
On the matter of drive size, I'm not running a testing lab, so I'm not yet trying to do simulation or replication samples for troubleshooting. But I was able to get 333 MB/s sequential reads and about 150 MB/s in sequential writes with an SATA-III SSD on an SATA-III controller to accelerate an SATA-III Veloci-Raptor -- both ported to the SATA-III plugs. To do that, I used the 20GB option to make the cache on a 120GB SDD, and used the remainder of the available space as a formatted partition. When I used the 64GB option, seq. reads benched at 252 MB/s, and the writes were around 133.
These results may be insignificant in predicting performance of SATA-III drives and connections against either an SATA-III or SATA-II HDD ported as SATA-II. Given the specs in sustained throughput for the Veloci-Raptor, I don't think it saturates the bandwidth of SATA-II. And in all cases, performance moves toward 80% of SDD standalone results, and "as much as" 400% of the HDD standalone benchies.
Under my assessment then, I think the Patriot Pyro 60 GB drive shows great promise. My tests were done on an Intel series 510 Elm Crest, with seq-reads/seq-writes 450/250. The Patriot is both rated and confirmed by customers at more than 500/400 [or thereabouts.] I just placed my order for one, so we'll see how it pans out.
You could use the entire drive as a cache, or try it as 20GB and 60GB to see what results you obtain. Right now, the Egg has them for around $99 after $20 rebate (and state taxes at the appropriate rate on $119.)
There is some "skinny" I heard that running too many disk benchies on the drive or especially its cache is not recommended. I suspect this may affect each subsequent result to its detriment. But it really depends on whether TRIM is implemented in the OS natively, as it is in Win 7. Probably best to benchmark right after creating a new cache.