I have to disagree with AndreasS and, I hope, help reassure Karsten. Of course it's possible to have a college degree and still use language poorly. Surely no one would claim that all German university graduates are competent judges of Rechtschreibung. Just as you wouldn't ask a philosopher to choose your cancer treatment, so you wouldn't ask a biologist to rule on grammar.
In my experience, educated speakers are, by definition, people who use 'criterion' and 'phenomenon' and would never accept, much less use, 'criteria' or 'phenomena' in the singular. (Unlike, say, 'data,' a far more borderline case.) I might even bet a virtual beer that trusted LEO users such as Peter <us>, Nancy, Oliver, RES-can, etc. would agree. (Where's Norbert, BTW?) Maybe at least a few other AE speakers will have time to comment.
It speaks volumes that even Webster's, which is normally well toward the descriptivist end of the spectrum, has seen fit to add a usage note, however tentative, on this word. Any hint of caution from them is probably equivalent to an outright thumbs-down from other dictionaries.
Speaking of which, American Heritage, which is generally more helpful than M-W when it comes to usage, also explicitly rejects the singular (see above). Note that AHD doesn't even cite usage poll results ('63% of the members of our usage panel find this usage unacceptable'), which they're careful to do whenever there's actually a significant minority opinion among educated speakers. So the implication is that their expert consultants are unanimous on this point.
If LEO were able to append the kind of usage note found in OED, AHD, M-W, etc., leaving in an entry such as this would of course not be unreasonable. (Even a one-word marking like [nonstandard], or a little gold 'u' button alerting users to consult a usage guide, would be better than nothing.) In the absence of any such capability, however, the least misleading option is surely to list 'criteria' only as a plural.