Fluctuation in Gate to Source Voltage

I have a 1.25 kW LDMOS amplifier. It works fine at low powers of about 100 W. But beyond that I see an increase in the display on the power supply for the Gate to Source Voltage and Drain current increases too. The CRO also shows a fluctuating pulse. The output power on the power meter fluctuates a lot too.
Following are the details:

Drain Voltage: 50 V
Gate to Source Voltage : 2.495 V (Increases to 3.8 V by itself in case of fluctuations)
Signal Duty Cycle of 10% (Pulse width : 1mS and PRT: 10 mS)
Drain Current Variation of around 1 A (from 2A to 3A) at maximum power.

If I continue increasing the output power, there is a spark and some smoke and I lose the transistor.
Please help!

First thing that have to look in those high-power failures, is the PCB layout.
Second have to look to the power supply (bias) filtering and decoupling, and any unwanted coupling between high power components.

It probably oscillates.

There could be some charge pumping / unanticipated
rectification of the Miller current back into the gate
at higher powers (meaning higher drain amplitude).
Is the gate really DC driven? DMMs and power supply
internal meters can be pretty bad at dealing with a
lot of AC content too. A 'scope shot might give you
more understanding, probe from supply to gate along
whatever chain drives Vgs. Also if source is "busy" be
sure you know whether Vgs is being determined by
gate node or source node excursion relative to the
measurement reference (and that you have the right
one).