Here's what I ultimately ended up doing after trying to exhaust your very helpful suggestions.
I used 5 Duplex sales. The only triplex or fourplex sales in the MLS (and it goes back to 2003) that were remotely comparable sold as packages with other triplex and fourplexes (6-12 units). I found no listings (active, expired, withdrawn) either.
I expanded to the only other MLS system within 50 miles.<>
It's time like that, when you need to swing by the county seat or call in, and get records for fsbo's, if you're lucky.
Step one is to observe the area. Visually troll the neighborhood and try to find one. Sometimes just having a similar build is a good supportive point, even if you can't comp against it. Also the counties approach to assessed value can provide ratios against duplexes, that may or may not be helpful, but hopefully will be.
Step two is to comp against dups. Nice job on that one. No big deal, because those sorts of units still appeal to similar buyer category, so they're fair comparables. 2, 3, 4, they're all the same, when it comes to some sort of lateral comparability. None of them are single family, but all of them are potentially mixed use for rental or owner and rental.
I've found personally, that the larger unit amidst smaller units, usually gets a proportional lag or diminished return on the ppsf sort of analysis for both mv and rent income. It depends on the build though. It's easier to comp them all out together when the 3 and 4 unit is really just quirky and perhaps does not have that much additional stature and footprint. Thereby justifying the appraisal position that cost to build was not equal. But if you've got a true sizing difference where they're like row homes, and the 2's are similar, but this one oddball has the extra 3rd same size unit, that's tougher. So you just extrapolate and appraise under the assumption of equality. The subject would therefore be same value as duplexes, plus the additional unit which is a simple extrapolation of average single unit value. So it would be 1.5x value. However, underwriting will probably get their socks knocked off and demand a bracketing unit. Sorry, only the ppsf analysis brackets under the logical premise of equality, and that's as good as it gets. I would not have reached out to another very distant area though. Extract, extrapolate, deduce, and turn it in. I've comped a du, against only sf's, because that's all the 'local market' had to bear.
Nice one.