Effects of impedance matching between 50 and 75 Ohm coaxial cables for 10 Mbit/s, Manchester-coded signals (20 MHz)
The refection coefficient due an impedance mismatch is: -
\$\dfrac{R-Z_o}{R+Zo}\$
Where Zo is the impedance of the cable and R is the source or load resistance.
And, for your 50/75 ohm setup will be -0.2. So the signal you put down the cable of (say) 3Vp-p will produce a reflection of 0.6Vp-p. Is this too much? It's not great but it's certainly not terrible.
Experience1 has shown that the resistive matching network is a good option for 10 base 2 Ethernet only at a first glance. It helps improve the situation when it comes to RF signal quality, but I have overlooked the issues caused by the way 10 base 2 handles collision detection, which are low-frequency effects and can be understood by simple DC considerations.
The connection will work best without any resisitive impedance matching network between the 50 Ω terminations and the 75 Ω cable segment.
Signal reflections and overshoots caused by the mismatch won't bother the transceivers much, but collision detection looks at the average (filtered) current into the cable, and with the resistive matching network, the current level is sometimes out of the specified limits. It all boils down to a consideration of DC currents created by the transmitters' voltages being dropped across the 50 Ω terminations of the cable (I=U/R). Adding thee resistive network will create a parallel path to the terminations and increase the DC current. This may sometimes mess with the collision detection. In my experience, this will mainly happen on hot summer days with high humidity levels, probably because of increased DC leakage along the dielectric in the coax.
TL, DR: 10 base 2 will easily handle the abuse of being sent over 75 Ω antenna coax. Overshoots, reflections, and any other side effects of the signal's RF part are not a concern. However, the collision detection looks at low-frequency currents, and it needs exactly two 50 Ω termination resistors at each end of the coax. Adding resistors will change the DC resistance of (50 Ω)/2 = 25 Ω and cause the collision detection circuits to work unreliably.
Having read around the internetsTM and having talked to some pretty experienced, old-school LAN experts has shown that this is a very common misconception. Therefore, please excuse the bold typeface font above. The misconception is even on wikipedia, as this related question shows.
Footnote:
1 Looking at the date of the original question, I have noticed that the system, with and without the resistive matching network, has now been in use for more than two years. I had trouble on some hot days in the summer of 2015. Then, I removed the resistive matching network and have had no issues at all ever since.