VARC isn't a particularly big problem.

100 * (ABS(L - L1) + ABS(L1 - L2) + ABS(L2 - L3)) / (ABS(L > L1) * (L - L1) + ABS(L1 > L2) * (L1 - L2) + ABS(L2 > L3) * (L2 - L3))

And you have come very close to what I had for an unexpanded VARA. Once we have VARA, we don't need anything above it. You appear to be missing the outer XAVG(,21) structure and I'm not quite sure if your factors or mine are correct (they don't match).

XAVG(XAVG((6.68*XAVG(MINL450,21)+6.64*XAVG(MINL250,21)+6.58*XAVG(MINL90,21)+3.464*(XAVG(MAXH450,21)+XAVG(MAXH250,21)+XAVG(MAXH90,21)))/36*1.738,21),21)

The problem is the nested 21-period exponential moving averages which go three deep. If the EMAs were different periods, then we could combine them in a way which does not require a manual expansion of the moving averages but that is not the case. This means substituting in the base items being averaged into something like:

(1/2)*(1-a)^3
*(1*2*C0+a
*(2*3*C1+a
*(3*4*C2+a
*...
*((k+1)*(k+2)*Ck+a
*...)))...

Where C0, C1, C2 and Ck are bars ago versions of MAXH450, MAXH250, MAXH90, MINL450, MINL250 and MINL90. And we need to take this expansion out to 100+ bars in order for the approximation to converge on an accurate value.

Cascades of Moving Averages

And the calculation for VARA needs to be included in every comparison used to create VARD for every term. The manual expansion of the 3-period exponential moving average can probably be done using 15-20 terms, but that still means 15-20 copies of bars ago versions of VARA to create VARD.

Then VARF requires getting the highest value for VARD over 60 bars. The technique I've devised for this has only even been expanded out to 32 bars because anything longer is too long and slow to be practical even when we aren't trying to find the high of something as complicated as VARD. Just take a look at the following topic to see how quickly the formulas increase in size.

Min Max PCFs

And then VARG itself ends up being a 3-period exponential moving average as well which needs to incorporate 15-20 bars ago copies of VARD and VARF.

We could do this using something with a full programming language like the RealCode in StockFinder or in TC2000 if the XAVG(), AVG(), MIN() and MAX() functions could accept complex formulas as arguments instead of being limited to only taking one of the indicators built into the PCF Language as arguments, but as things stand, it is not possible to create a PCF for VARG in TC2000.