What is the best option to solve too wide quarterwave transformer problem?

I need to match 16 Ohm MMIC output to 50 Ohm microstrip line at 24GHz.
Currently I built a simulation where Sqrt(16*50)=28 Ohm quarterwave transformer is used.
The problem is substrate is relative thick with high dielectric constant (t=0.5mm Er > 4). This resulted in quarterwave transformer sizes where W>>L (width is much larger than length). With all optimization best result for S11 is -3dB, and obviously it does not work as quarterwave transformer (line is too wide).
I tried to use more common substrates like RO4350, and results are good (S11 is less than -10dB), Used Er=3.6 and thickness of 0.25mm.

Currently I have two ideas, but open to more.
Here is my ideas:

1) instead of matching 16 Ohm to 50 Ohm , perform matching of 16 Ohm to much higher "virtual" impedance (100 Ohm, 200 Ohm etc.). Then, put a second quarterwave transformer, to match say 200 Ohm to 50 Ohm. Does it make sense?

2) try other ideas, like twelfth wave impedance transformer, not sure if it is good idea for microstrip.

try coplanar waveguide instead of microstrip?

also you can play some tricks with even and odd mode impedances to achieve a better aspect ratio for the quarter wave transformer

(Matheii young and jones, page 215)

might even be able to do some lumped element loading of the line to achieve a lower effective characteristic impedance

if it is a narrowband match, you could just resonate it.

oh, also you can do a 3/4 wave transformer, instead of a 1/4 wave one. it will be a narrower bandwidth match though

You can use tricks to make your line more capacitive, to lower the line impedance. If you have an EM simulation tool to verify the result: place a top level ground near to the signal line, connected by vias to the bottom ground.