Any ideas?

I have a circuit as follows

9 volt dc input, 555 oscillator driving 1:500 coil, bridge rectifier and
smoothing caps, giving a rail of 4.5kv. From this rail, 3k3 in series with
33nf cap to ground. Spark gap in parallel with this cap. All works fine and
runs for 20 hours in two or three 8 hour sessions. Then at switch on, some
kind of plasma / corona effect at the wire exit of the diodes, followed by
further breakdown. Diodes started with a Vrrm of 5kv, 5ma, and have been
upped to 30kv, 3ma. The same problem occurs, but now takes a little longer.
Anti corona varnish does not help.

I need to run the unit for sessions of 10 hours, and ideally continuosly.

Any ideas??

Nick