The switch fabric provides the bandwidth between ports. To guarantee all ports can pass full line rate, you would need 2x ports bandwidths (assuming duplex ports), so for example, for 8 gig ports you would need 16 Gbps bandwidth.
Many switches do not provide enough bandwidth for all their ports to run concurrently at full capacity. If the switch doesn't, the switch will delay or drop packets when the fabric bandwidth is exceeded.
Besides raw fabric bandwidth, architecture of the fabric can also limit performance. Sometimes multiple ports share bandwidth to the main fabric and/or queuing is done on ingress to the fabric rather than egress. This latter situation can lead to head-of-line blocking (i.e. one busy port blocks other ports). ([edit] head-of-line block example - real traffic - vehicle ahead making left [and stopped due to on-coming traffic], but you can't pass it although road ahead is clear)
The impact of what's described above, to forwarding performance, can be very similar to the impact of a switch uplink. If you have an 8 port switch, all same port speed, and use one port for an uplink, all the other ports can bottleneck on the uplink. If the 8 port switch had a 10x faster dedicated uplink, those ports would not bottleneck on the uplink. If instead of 8 ports, you had 24 ports, but again a 10x uplink, you can still bottleneck on the uplink, but perhaps not as often as when there were 8 same type of ports because the bandwidth ratios between edge ports and uplink is better (7:1 vs. 8:10 vs. 24:10). It's similar for fabric bandwidths.
