When power is first applied the voltage at connector 1 of T2 is first controlled by R1/R3 (47K) and not by C3/C4 as these low value capacitors have very little effect at 60Hz, they are selected to suit the transducer output frequency. 

As the voltage raises, it only needs to get to 1.2Volts before Q2 starts conducting and this in conjunction with the low value of the capacitors mentioned above causes the voltage at connector 1 of T2 to go much higher towards the positive rail than the mid point it is normally at during operation. 

This means that during startup of the oscillator, Q1 will be sinking from a much higher voltage than Q2. As power output is  V^2/R this means there would be a significant spike in power output at startup without R4 limiting the current through Q1.

Also, it looks like you have an error in your schematic on connector 1 and 2 of T2.

I agree that the transformer looks funky - they're using it as an inductor.

I've seen transformers used as inductors before, typically the unused side is shorted and grounded. The circuit is consistent with this theory, shorting and connecting to the middle of the voltage divider.

I buzzed out the transformer and double checked the PCB trace - you can see the traces in the X-ray view on the build log post (link below). One side is completely shorted, and the generated signal goes through the other winding. The transformer can be tapped in one of two places to allow for different inductors for 40kHz and 28kHz operation.

https://hackaday.io/project/4689-improve-the-bosch-haber-process/log/16375-reverse-engineering-a-cheap-ebay-ultrasonic-power-supply

BTW, thanks for the response.

Looks like a current limiter to me.  It limits the max current through the transistors.  I don't think your connections to the transformers are correct.  The top one appears to have the left side either open circuit or shorted.