As I understand it, the 3 settings work together like this:
(first of all, I read several statements in vSphere books that this setting isn't used very often, only in very particular situations, so not something you would usually configure?)
1) you start by configuring the Average bandwidth, as a way of limiting the bandwidth, used by the VMs connected via the vSwitch or (mostly) portgroup, to a value lower than what the physical uplink adapter allows. Can be used for example to protect VMs on another portgroup that you consider to have higher priority for bandwidth, to be bandwidth-starved
2) the "burst size" setting is an optional additional setting that allows you to further refine the behaviour: if the bandwidth used by the portgroup consistently remains under your configured Average, you can sort of "reward" it for good behaviour by allowing it to build up a buffer/bonus of extra capacity that will be made available to it for consumption in short bursts when necessary. The "burst size" is the size of this bonus in KB that can be built up.
3) further refinement is available through the "peak bandwidth" setting: this one will limit at which rate the portgroup is able to consume its burst bonus when it makes use of it during a burst.
