Colorspace, as Chaboud mentions above, is a term that describes the concept of individual colours as coordinates in n-dimensional space, where N would be the amount of component channels in that colour format. L*a*b* and l*u*v* are useful colorspaces because of the ease of editing luminosity seperately from different chrominances, or colour and saturation.
The concept of perpetual uniformity means that to the human eye, the colours will appear rise in one component linearly, whereas the digital value of one attribute may rise exponentially (or logarithmically). This concept is employed by gamma, which allows the RGB luminance to be stretched nonlinearly for perceptual uniformity. Other formats which make use of a larger or different colorspace for holding useful informations are HDRI or other variants which make use of 32-bit or 64-bit colorspace. RGB traditionally has a 24-bit colorspace divided equally into red, green, and blue components of 8-bits each. This allows for a total of 16777216 colours, which is slightly more than the human eye can distinguish. 32-bit or 48-bit HDRI images make use of the extra colorspace for holding contrast and lighting information, based on the principle that whitepoint is a variable with any conceivable value. Where RGB will cut off incredibly bright sources at the peak brightness of (255,255,255), HDRI can hold any value for whitepoint, blackpoint and exposure, allowing for a more dynamic range of colours in an image. An example of where this is useful is for radiance maps, used to light up scenes in computer animation: in a scene with the sun and a candle, both light sources may appear to be clipped off at white, but they each emit varying amounts of light, depending on the dynamic luminance value assigned to each pixel. Radiosity and other image based lighting solutions take advantage of this to get higher realism.
It should be noted that the concept of colour is isolated to only humans and animals with the ocular and mental capacity for it. The cones in your eye detect different wavelengths of light and assign a colour to them neurologically. In that way, the perceptual colorspace of a human being is an "imagined" value, and is only a way to visualize and distuingish varying wavelengths of light. Because we can only detect or distuingish a limited spectrum of light, the chief advantage of a digital colorspace larger than the human perceptual colorspace is for holding additional attributes of the light, apart from the wavelength, such as the luminous flux. Although 24-bit is more than enough colours for displaying an image, there are many benefits of a large colorspace or differently modelled colorspace. Non-linear CYMK colour is useful for printing because it is subtractive: the greater the value of a component, the closer to black the image becomes. It is not perfect however, because printed ink cannot represent the full spectrum of visual colours. L*u*v* is very useful, although uncommon, because it allows for incredibly high precision images in a very conveniant format which allows luminance to be seperated from saturation and hue. One of the highest precision and most accurate colorspaces is the one used by CIE XYZ, which is another node all on it's own.