The only relevant rule is how it was spelled in the original description - if it was misspelled by the describer, it is nevertheless the correct species epithet. Would you know what the original spelling is, @borisb?

There's tons of cases where describers misspelled the taxon name - always worth checking the original description first before assuming anything.

I should add one thing: the gender of the species epithet has to agree with the one of the genus - if that's not the case, the species epithet is changed accordingly. Not applicable here, though.

Yes, it's called noun in apposition (I didn't mean to go into the details of the zoological code).

So what should be done here? The original spelling is Coccinella 14-guttata, and the description says "punctis quatuordecim [sic!] albis", which however is irrelevant in terms of nomenclature, so Calvia quattuordecimguttata should be the correct current combination.

See also www.animalbase.uni-goettingen.de/zooweb/servlet/AnimalBase/home/speciestaxon?id=42297

Personally, I think this swap has gone the wrong way.

The code itself gives absolutely no reason to correct spelling in such cases., and it is clear from Latin dictionaries that the spelling with one 't' was always a variant. Since that spelling was the one that Linnaeus used, I would have thought there should be a bias to accepting it for the expansion of the numeral. Article 32 doesn't address this. For what it is worth, Catalogue of Life includes twice as many scientific names with the spelling "quatuor-" as with "quattuor-" (97 vs 48):

http://www.catalogueoflife.org/col/search/all/key/quatuor/fossil/0/match/0
http://www.catalogueoflife.org/col/search/all/key/quattuor/fossil/0/match/0

There are nearly 80 times as many Google hits for "Calvia quatuordecimguttata" are for "Calvia quattuordecimpunctata".

Not exactly. Both the Latin and Greek words derive ultimately from an Indo-European root which must have started with some kind of voiced labiovelar stop which gave rise to the Latin by loss of voicing and the Greek by a more brutal reduction. It certainly seems that double-t was the more common (or preferred) spelling. However, the single-t version is certainly there in Lewis and Short and elsewhere. More importantly, even if this represented a spelling mistake, that is not a reason under the code to correct it. The fact that Linnaeus used the 14-guttata spelling (which the code now requires to be spelled out) might have left the choice open, but I am certain the spelling chosen in the description and the fact that the Fauna Svecica used the single-t version is ample precedent for deciding against the double-t in this case, quite beside the fact that the double-t spelling doesn't really seem to be in use anywhere authoritative outside the UK.