With the 10k fixed resistor, there are some clipping on the negative cycle and oscillations near the crossover that show up in my simulation. With that reduced to 2.2k and the 0. 22uF cap made smaller (22nF); I get 4.5W rms audio at less than 0.1% THD with a 1kHz test input. It seems to work well from 20Hz (THD 0.3%) to about 10kHz. At 10kHz, distortion (THD 0.4%) starts rising up to 2.5% at 20kHz.
This is the result with a pure resistor load. Performance with a real speaker load can be very different with other unexpected surprises.