Parallel Transistors in Constant Current Load
Directly paralleling the transistors is a terrible idea- the threshold voltages of the MOSFETs will be different and one transistor will take much more current than the other- further you are only measuring one current so the total current may be greatly in error. It's worse than with BJTs.
Fortunately op-amps are cheap and you can just use two or more (one per MOSFET), fed from the same control voltage, each fed back from an individual source resistor. Use a single supply op-amp, not an LM741. That way each current will be precisely the desired fraction of the total, and the total will also be accurate. There is no need to match transistors or resistors (but the resistors should be accurate enough for your desired current accuracy). The currents from each transistor will simply add.
You are right to worry about that power level in an MTP3055, but it's nowhere near the limit of the package. For instance, the Vishay IRL540 is a TO220, and it's rated for 150 watts. You just need to make sure you've got a good heat sink. And by good heat sink I mean a big honker. No clip-on jobbers.
Assuming you still want to worry, your alternative approach will probably work. It's true that one of the MOSFETs will be running open loop, but I'd guess you'll be all right. Matching the transistors would be a good idea, but I doubt it's necessary. Use 1% resistors. Isolating the gates from each other with about a 100 ohm resistor for each gate is also nice.
The ultimate way to go if you insist on using wimpy MOSFETs is simply to duplicate your load circuit, with each load set to pull 1 amp. Put them in parallel and you're set. Op amps are cheap, although I'd strongly advise upgrading from 741s. At the least, use single-supply op amps and save yourself the trouble of providing dual supplies - and with 741s you really need dual supplies.
I would use a separate op-amp for each switch, each with its own resistor so that you have multiple independent current control loops in parallel. The general rule of thumb is it OK to use MOFSETs in parallel because as they heat up their on-resistance increases, and so they generally share current well, but this is for use as a on/off switch. When used in an analog fashion such as this, I'd imagine differences between components could result in very different currents through each transistor for a given gate voltage.
You could use bipolar transistors with a similar circuit as long as the op-amp output current capability is sufficient for whatever the transistor gain requires. You have to be careful using MOSFETs for analog circuits like this because some (especially trench types) are designed for fast switching full-on or full-off type applications and fry quickly when left in-between too long with significant current. This MOSFET isn't of that type but check the "Maximum Rated Forward Biased Safe Operating Area" graph in the datasheet. Even with two in parallel you are dangerously close to the edge of the safe operating area. I'm not a transistor expert but typically find BJTs more robust for analog circuits (I've killed many more MOSFETs than BJTs).
For projects like this I think one of the biggest issues you need to think about is how you are going to dump heat. I think using multiple transistors is a great idea in addition to big heat sinks, fans, etc. Junction to ambient thermal resistance for T0-220 packages is usually around 60 degC per watt without any heat sink. So 22W would rise the junction temperature of the transistor 1,320 degC over ambient! That is definitely going to release all magic smoke! You usually need to keep the junction temperature below 150degC or so depending on the part.