We perform the multiplication primarily for practical reasons, namely to make the demodulator much simpler and cheaper. Consider this. The obvious way, in theory, to get modulation out of a signal in a digital receiver, is to sample the incoming RF directly and then do digital signal processing on the samples to pull out FM or QPSK or any kind of modulation. The trouble with that idea is that, for most RF communications, the sampling rate would have to be so high (remember Nyquist) as to make the hardware extremely expensive or maybe even impossible.
Better to mix the RF down to baseband so that sampling can be a lot slower and a lot cheaper. But you can't just mix it with one mixer because then you lose some of the information (some of the modulation gets messed up). The way to preserve all the information (that is, all the phase, frequency and amplitude info) is to use the IQ approach. With the two paths of mixing, using a 90 degree phase offset between the two local oscillators guarantees that no information is lost. That is the point of IQ downconversion or IQ demodulation.
Your question, referring to a PLL, implies that you may be confusing IQ downconversion with another thing called a quadrature demodulator. These are not the same thing at all. The quadrature demodulator is an FM demodulator and can be directly compared to a PLL for FM demodulation. A PLL is not able to demodulate any amplitude modulation from a complex waveform, so it is not as capable as IQ demodulation.