Why does bootstrap should be an alternative to isolation ? The issues discussed above are related to the lack of isolation between control and power, and I believe they do not apply to the bootsrtap technique.
The major drawback of bootstrap is simply related to the limitation in the possible PWM strategy: in practice there's an upper boundary to the on-time of the high-side switch, that depends on the discharge time of the bootstrap capacitor, and a lower boundary to the on-time of the low-side switch, that depends on the charge time of the same capacitor. Typical PWM strategies in
VFDs (like sinusoidal or space-vector) normally have no problem with bootstrap. Earliest bootstrap drivers used to suffer from latchup phenomenon because of negative voltage on the GND pin caused by high current transients in the parasitic L of the layout. But newest devices (like IR21864) with some external components for protection (as suggested by app-notes) and a well-done layout will perform very well up to tens of amps and hundred of kHz (and if the driver max current is not enough to drive a large gate, you can always, just put a couple of transistors in totem pole configuration).
Isolation is another issue: high-current, high-voltage and fast transients management may take big advantage from isolation. So why not considering to use bootsrap technique with isolated commands ? That's what we normally do, when possible.
The major advantages using bootstrap are:
* no need to design and prototype a custom flyback transformer with multiple outputs
* much easier layout (no need to wire the flyback isolated outputs to the high side switches)
Instead cost reduction may be only marginal...