By leaving the negative lead of C26 disconnected, you are effectively disconnecting the reservoir capacitor. The diodes are then seeing a higher impedance load, as the choke L17 appears as a high reactance to the ripple component of the full wave rectified pulses of DC coming from the bridge rectifier.
The diodes you have used are rated at 1 amp, and that's an absolute maximum DC current. When forward biased, each diode will drop about 0.6 volt so it will be dissipating 0.6 watt, and these diodes are physically small, so they'll run warm. The current pulses into C26, however, will only last around 2.5 milliseconds per 10 millisecond half-cycle, and in that time the diodes have to pass enough current to recharge the capacitors to their peak voltage. So if the average current drawn from the smoothing network C26-L17-C25 is 250mA, the peak current through the diodes will be seven or eight times that. In other words, your 1N4007 diodes are operating above their absolute maximum current ratings.
I would suggest using something like 1N5400 3-amp 50 volt diodes or an encapsulated bridge rectifier like
this one from Maplin. You don't need a high peak inverse voltage rating, so your 1kV would be overkill, and 50 volts is more than adequate as the actual PIV across the diodes will be less than 5 volts.
Incidentally, you probably paid well over the odds for branded electrolytic capacitors - they seem to be targeted at the audio market, where high price is deemed to equate with good quality sound - and any normal, decent capacitor would be electrically identical.
By the way, why did you replace the original metal rectifier in the first place? Had it actually failed? If not and you still have it, you could try putting it back...