As previously mentioned, the standard Internet MTU has become 1,500 bytes - the Ethernet MTU.
Due to the known flaws of Path MTU discovery, and to the fact that most Internet paths traverse an Ethernet network somewhere on the way, most vendors disable Path MTU discovery on their hosts, and set the default MTU to 1,500 bytes.
Apart from significantly downgrading performance when connected to higher MTU networks, other problems arise with today's high speed networks.
At Giga-bit/sec speeds, the amount of CPU utilization and network overhead required to achieve a desired throughput are enormous. Since smaller MTU means more packets, the CPU encounters more interrupts per second, and the amount of non-data transport (namely, the headers) increases.

Many proposals are being submitted these days, to the standardizing organizations, in hope of deploying large MTUs in the high performance parts of the Internet.
The following proposal, by Matt Mathis (www.psc.edu/~mathis/MTU) is based on the belief that a factor of ten bandwidth steps should impose a factor of 8 increase in payload size, and a 20% reduction in packet time :

The above numbers are only approximate and are likely to evolve with better understanding of how MTU affects the balance between switching costs and end-system costs vs end-to-end performance. Nevertheless, they encapsulate the desire to optimize network utilization by enlarging the MTU.
These motions to enlarge MTU, should also be accompanied by improvement of the Path MTU discovery algorithm, in order to achieve full optimization. Many proposals for new Path MTU discovery algorithms have already been submitted, and an example of such can be seen at www.psc.edu/~mathis/MTU/draft-mathis-MSS-discovery.txt