Yes you can and I've done with lower power OPAs (in order to preserve low board height etc.) The main gotchas are:
- You need a way (e.g. series resistors) to limit Voffset induced currents circulating between amps. Series resistors will blow away some power. Their value needs to be evaluated carefully. The higher they are, the more protection the amps have, but the more power you blow away needlessly. As @pjc50 correctly notes below the 10-ohm values shown in the circuit from the OP551 datasheet below are rather unsuitable for higher-power amps (OPA551 only goes up to 0.25A in "normal" use so loses in those resistors are about 0.6W each in that case. For 10A at with 10-ohm resistor you'd blow away a [theoretical] 1kW!! (In reality, the amp's output will likely saturate [hit rail] well before that as few amps can output 100V). Apex Microtechnology, which sells mostly higher-powered amps, recommends in their appnote (see link at end) only values in the 0.1 to 1 ohm for these series resistors (corresponding to power losses of 10W to 100W in the 10A scenario.) It so happens that the OPA541 actually has such a circuit in its datasheet (with a rather strange label, and it is not really being discussed in the text--hopefully they revise that) with a master/slave config and using 0.1 ohm series-resistors.
(The capacitors in the latter circuit are there for stability improvement, almost certainly, see below. The odd looking connections from the output to what might be misinterpreted as a power rail of each opamp are actually going to the current sense pin[s].)
One more slight variation on this theme is found in the OPA549 datasheet. In this (also high-power, 2x10A) case they also use only 0.1ohm series resistors but they close the master's feedback loop locally, probably because the small values of the series resistors don't cause a significant voltage drop.
The slew rate will go down (at least if you use the cascade config like all the above circuits do) but in your case [50Hz] you probably don't care about this in the slightest. You'll obviously need two slaves to get 30A in your case. The slaves don't need to be cascaded relative to each other. Regarding topology, some Intersil engineers say in their AN-1111 you're better off with a non-cascaded topology to get better slew rate. You'd need a few more passive parts for that and in my case I didn't need the improved slew rate, and you surely don't either.
Stability is worse if you change the feedback loop to be global, as done above with OPA551's. However for higher values of output resistor that's preferable because the global loop automatically deals with (for gain-setting purposes etc.) the voltage drop across the series resistor[s]. I've had to add some compensation capacitors that I didn't need with a single amp in order to deal with some extreme input cases that caused oscillation in the parallel config. But my inputs were 100kHz in some cases (and probably had higher freq noise on them too).
You may also need to worry about which opamp is going to limit first (temp and/or current) since you're planning to max them out (I wasn't).
Other resources worth reading on the topic are this TI blog post and the APEX note AN-26; the latter details more advanced issues like stability/compensation.