Well, here's one idea. Diodes D1-D4 create a full-wave rectifier, meaning for the most part--only two diodes are conducting ever at any given moment. IE: When D1 and D4 are on, D2 and D3 are off and can be ignored.
If you take that, plus the fact that the rectifier's output is roughly ABS( 220cos(2π*50*t)), you could redraw the circuit this way:
**broken link removed**
Looking a the largest circuit mesh:
Vi = Vd4 + Vr1 + Vc1 + Vdz + Vd1
RODALCO can help you set R1 and C1.
There are still 2 more circuit loops that need to be accounted for (between DZ, R2 and C2), but since the Zener diode is basically being used as a voltage regulator, we could assume that Vdz never exceeds DZ's zener voltage (Vzener). So if we ignore C2 briefly, the voltage on the zener will resemble a clipped sinusoid...
if: Vi - Vd4 - Vr1 - Vc1 - Vd1 < Vzener (condition A)
then: Vdz = Vi - Vd4 - Vr1 - Vc1
otherwise: Vdz = Vzener
To choose the value of C2 that works with R2, you have to a choose an RC time constant (R2*C2 = X seconds) that exceeds the amount of time "condition A" above is satified. Looking at it that way, I don't think there's any upper limit to a good C2 value, just the lower limit that's imposed by condition A.
That's just one possibility.