With a bit more searching I think I may have come across the solutions.
It looks like, for a simple combination of inputs a simple resistive circuit as shown in the question with simple mixing of the outputs is fine, if gain control is required on the individual inputs then it is better to use a virtual earth mixer is best.... so I think what I propose should work.
I'll leave the question open for a couple more days to see if anyone can add any further inputs.
Yes that solved it. I have three pots dividing the voltage across the three speakers (left, right, centre) at the speaker outs on the receiver. The outputs of the pots are combined (mixed) into one input for the mid-woofer crossover and then through to the amp... pretty much perfect now... one day (when we get a bigger home) ill be able to get proper speakers and get the best sound out but this is a great workaround for a small house.
The classic balance control is shown here. This needs to be driven by a relatively low impedance output, and fed into a high impedance input.
simulate this circuit – Schematic created using CircuitLab
R2 and R3 are equal. You choose their ratio to the resistance of the linear control pot R1 to choose the slope of the gain control in the middle region of rotation.
Here is a plot of the dB gain of each channel, for selections of R2 and R3 of 3k, 10k and 30k. Obviously as they get bigger compared to the balance pot, there will be a larger range of control around the balance point, and also more loss at the balance point, also known as gain boost at extreme rotation.
As you can see, with a 100k balance pot, there is less than 1dB loss at balance when 3k is used. There's about 2dB loss using 10k, and about 4dB loss using 30k.
With those 3 curves plotted, it's easy enough to interpolate to intermediate values, and even extrapolate to what would happen with more extreme selections. However, when adjustable balance is needed, most audio users seem happy with a control somewhere between the 2dB and 4dB curves.
If you are not happy with one of that family of curves, then I suggest you sketch out a gain control curve you would be happy with, gain versus rotation, in the format shown above, add the sketch to your post, and we'll see what we can do.
I've used dB in the above graph as most audio engineers use those. 3dB difference sounds the same, whether it's on a loud or quiet signal. Linear units don't behave like that. When you find a software volume control in a media player that goes from too quiet to OK when going from 5 to 10, and then doesn't seem to get much louder going from 20 to 50, you've found a linear control implemented by a programmer with no prior audio experience. There are some products by surprisingly high profile brands that still do this.
It's easy enough to switch between dB and linear units. A dB gain is 20*log10(linear_gain). The linear gain is 10^(dB_gain/20). In very round numbers, -2dB is a gain of about 0.8, and -4dB is about 0.63.
here's the response I got after emailing Diodes:
You cannot bridge the PAM8403 (as you can't bridge two pairs of differential outputs), and Diodes have recommended against using the two outputs in parallel.