ReadIris has done a journeyman (i.e. not spectacular) job for me except for a few issues.

First, the 50 page limit is a pain, and they send too many promotional emails. But given that...

Its ability to learn and improve seems to be poor. If I partially OCR a document and correct it as it goes, then abort it and start again, then it surely remembers the corrections I made -- basically I only have to continue my corrections from the point where I stopped. However, this learning appears to be so supremely specific to the particular OCR'ed glyphs that even if I save that dictionary and reload it after opening different pages from the same scanned PDF, it's as though ReadIris has had no training whatsoever. e.g. it again confirms with me what the numeral "3" is, and similar basic questions that I've definitely covered before.

It is apparently pathologically unsure of its recognition of punctuation. I have been asked to confirm quotation marks, dashes, periods, semicolons, colons, and parentheses more times than I can count. And even though it's correct 90% of the time, no matter how many times I say "Yup, you got that right," it still asks me 300 more times. I'd much rather correct the occasional mistake when I proofread the document after the fact, than waste my time having to go through this confirmation step endlessly.

If you load a document, OCR it (theoretically training ReadIris), then load a second document, your previous correction dictionary apparently just disappears and it starts from scratch. Of course that's not much of a loss, given my comment above, but that's a pretty nasty user experience.

I won't even mention that it just crashed on me today on the final page of an OCR job (oh wait, I just did... Never mind!)

Basically if I hadn't already bought it (over a year ago), I'd spend the $70 to upgrade my DevonThink Pro to DevonThink Pro Office to get the same (IIRC) OCR engine with a less-painful interface, by a company that actually speaks to its users.