If you don't actually need to know which is which, the smallest value (hottest, assuming these are NTC) dominates, which can be handy for a crude protection system.
Otherwise, you're asking for some kind of mux, whether it's an explicit digital one or not. For example, your voltage selection proposal is equivalent to sending that voltage to an ADC, and using the digital output to set a mux. Whether there's a worthwhile simplification with discrete parts, who knows, but you will find it very, very difficult to compete with, say, an 8-pin MCU.
Or to put it another way, you're proposing a mux with a multi-nary logic level selector input. That is, instead of binary 0/1, the voltages 4 to 4.99 = 0, 5-5.99 = 1, etc.
"Bandpass" could be used, if each thermistor is driven with a voltage at a different frequency, and the resulting currents are summed into an output. But then you need to do signal analysis to tease them apart, and you're still looking at an MCU as the best option (ADC + FFT + peak tracking) but with a hell of a lot more steps!
Which still isn't a terrible method, actually, but with a lot more considerations, and really only for applications that absolutely demand going to such serious extents. The telephone system went through just such a process back in the 1930s: frequency division multiplex, where individual phone channels are modulated into separate frequency channels, and all those channels are transmitted at once down a single wire (instead of dozens of wires -- a huge savings at a time when most telephone lines were overhead messes!).
(But yeah, unless there are really specific requirements, probably not necessary to do that here.)
So, really what you need is to step back and consider what IO pins (and types, digital/analog) you have available to perform the selection, and how much hardware you can afford (due to space, cost and design constraints) to implement it.

Tim