As suggested in theory they should be the same, in practice the amount of RF power actually output is not linear to the amount of consumed current for typical devices. Above 10dBm a typical RFM98 will consume 5mA per dBm, but at 2dBm its up to 16mA per dBm.
Then there are practicalities, using a bandwidth below 62.5Khz can be problematic, transmitter and receiver might be outside of the capture range due to natural manufacturing tolerances in the timing crystals used.
Increasing antenna height is often the easiest way of increasing range.
Great point about the TX power being more efficient at higher levels. I’ve also kept BW at or above 62.5khz for the reasons mentioned. So is there any reason you’d choose 62.5khz SF7 over 125khz SF8 or vice versa?
Well, the TCO might help the temperature drift, but it does not necessarily improve the accuracy of the tx frequency in relation to the rx frequency. Its not difficult to add a calibration offset in any case.
I have not had a a problem myself with long packets (7 secs or more) at BW62500, SF12
Getting packet failures with longer packets can happen if the optimisation bit is not set correctly.
I agree with @BoRRoZ : provided you are already using the maximum tx-power out of your node (and maybe even running in the the regulatory power limit) and the maximum SF, then it pays off to use a good antenna, maybe directional, and choose the best antenna position.
Remember that a good antenna helps you both in transmitting and receiving, whereas increasing TX-power only helps you in transmitting!