In a recent 1938 game I played as the US, Japan was rapidly kicked out of China on all fronts, until it managed to slow down the advance in the very north (holding only the port in Shanxi). Still, three years later, in late 1941, they DOW:ed the US. But since they had practically no army left, all they could do was to patrol the seas.
In a current 1936 game as Germany (it was easier to see what Japan was doing being in the Axis), they almost managed to conquer China when Shanxi fell. An sea invasion was, however, performed earlier in Shanghai. The Chinese redirected forces fro Shanxi and almost beat back the invasion, but then Shanxi fell, followed by Com China, after which the Japanese fanned out.
It seems to me that these two examples illustrate that fundamentally China is already too strong for the force that Japan deploys, but the Japanese can succeed depending on various (semi-)random events. The nice thing is that in FTM Japan actually seems to care about its strategic goals.
This leads me to believe that the problem is not in the balance. At least the 1936 scenario is quite balanced. The problem is that the balance is not a stable one, and very easily tips into the favor of one of the parties.
A nice addition to help with this (and other event triggered wars) would be to add an "anticipation" period, meaning that the AI will decide on pursuing such an event, it would deploy its forces for attack. This could be achieved by adding a "war_trigger" flag to political decisions. One could also refine the system of war goals by, for instance, adding priorities and weights. This would create a less random attack strategy, and Japan would have to deploy its forces accordingly.
On the defense, one could, as has been suggested earlier, rework the surrender of Shanxi and Com China, and perhaps introduce a "bitter peace" type of event. However, one could also use a more refined war goal system to emulate the historic situation. Upon achieving its high-priority goals, each side could go into a defensive mode, which was essentially what happened historically. I fact, there was very little actual fighting in China after the initial invasion and the final desperate offensive in 1944. This is very difficult to recreate if both sides are doing whatever they can to destroy their opponent, even if one removes some of he randomness of how they go about doing it.