Germanium...some added info for you.
I have not worked with opamps for a long time and in my day the LM725 was the hallmark of opamps. It was an instrumentation opamp meaning it operated very precisely. However, to get that precision, you had to add external circuits.
The X-Fi uses an NMJ2068 dual opamp, meaning it squeezes two opamps into one package. With a + and - input and output on each opamp, plus the required power supply inputs, all 8 pins are used up.
The LM725 was an 8 pin device as well but had only one opamp in the package. It had extra pins available with which external circuits could control the offset drift and supply frequency compensation.
If anything, the dual opamp used in the X-Fi suffers from a lack of that external control.
You might be able to remove the NMJ2068s and use some kind of daughter board to add 8 pin single opamps that allow external control. You would have to extend the NMJ2068 pads to the daughter board for input and output. You could take the required power supply voltage from elsewhere, or from the same pads.
If you get a data sheet for the NMJ2068 that shows an equivalent circuit, you will see the output being taken from between two resistors in between two output transistors. They are using the NPN/PNP complimentary symmetry arrangement which is a Class AB1 design. The AB1 has an inherent distortion at the crossover point between the +ve and -ve going output signal and it requires an output capacitor in some cases to smooth out the distortion.
The capacitor works much like a power supply capacitor, although it is actually coupling signal to the output. In a power supply, the output of the rectifier stage is a pulsating DC with the capacitor right across the stage output. The capacitor charges to the peak voltage of the positive DC pulse, but as the sinusoidal waveform of the DC pulse decreases, the cap tend to hold the voltage at the peak. Between pulses, the cap tends to hold the voltage flat causing a pure DC but it does drain off through the load, depending on the RC time constant.
When viewed through a scope, the DC has a ripple in it at either 60 hz or 120 hz. All RC-based circuits leave a slight ripple which can be removed with a regulator.
At the output of an audio amp, the cap is attached in series with the load to the neutral point between output transistors. The resistors you see on the equivalent drawing are the emitter resistors for each output transistor. As the input signal varies between +ve and -ve peaks, each output transistor will be on an equivalent amount of time. There is a point, however, between the +ve to -ve output signal transition (crossover point) where both transistors are off. In this region, the outputs from each output transistor do no coincide, causing a potential overlap which can be heard as a buzzing sound.
There are several solutions attempted but one of the simplest is to include an output capacitor. The cap can stabilize the crossover distortion region by maintaining the voltage till one transistor begins firing again. There are purely direct coupled output designs with the speaker connected right to the crossover point. Most amps I have seen, however, use a cap. If you omit it, you need to compensate using feedback and/or bias adjustments on the transistor input. Since there are none available in a dual opamp package, you can't get at them.
You would save yourself a lot of grief by leaving the caps in the circuit.