Try splitting it into two back-to-back at 2:7 and 7:20. (both as close as possible to 1:sqrt(10), which makes each reduction use very similar sized gears, but also picked for a reason which will become obvious below).

Not sure exactly what this is going into, but for funsies, you could even make the input/output axial by making the tooth count for each pair equal, as per the following working:

• LCM[ (2+7), (20+7) ] = LCM[9,27] = 27 (use this as initial tooth count for each pair)

• 2:7 * 3 = 6:21 (total 27)

• 7:20 * 1 = 7:20 (total 27)

• Let's multiply by 2 to make that 6t gear turn into a 12t: 12:42 and 14:40

• For a module 1 the PCD (pitch circle diameter) is m*t = 12, 42, 14, and 40mm. OD of the gear is usually this +2m, so 14,44,16,42mm.

• The total largest linear outside distance will be (t1+t2+2)*m = (54+2)*1 = 56mm

• Axle spacing should be (t1+t2)*m/2 = 54*1/2 = 27mm

Thus, by splitting it into two you can do a 1:10 ratio in the same space as pirvan's M0.5 gears, with the nearly-printable M1 tooth size, the input and output coaxial, the smallest gear is a respectable 12t, and the helpful/conventional property that each gear pair shares no prime factors in the tooth count (other than 2) which makes wear very even!

The smallest usable gears I was able to print were with a 1 module.  That would make your large gear 100mm (~4") diameter.

Here is a drawing of a gear set with .5 module

Edit:
P.S.  If you do want to try and print them, I can make and post the STL's, I just need to know the face width (the thickness of the gears), and if you want a hub (so you can use a set screw).