At the risk of incurring the wrath of the Internet, I'm going to disclose this right up front: I have not read all 15 pages of posts on this forum. No apologies.
(It might be time to summarize the posts-to-date in the wiki and come up with a list of side projects or volunteer tasks related to the GPIO interface...)
And now, I'm going to try to introduce what I hope is a new idea (on p16 of this thread, where it might never be seen):
There was a lot of discussion early on of multiplexing the GPIO pins.
If you want more GPIO, consider something like a Cypress EZ-USB chip and creating a lightweight USB peripheral. This is, more or less, how the industrial data acquisition products from Measurement Computing and National Instruments are built. Only, we should be able to do it for a fraction of the $300/board (and up) prices that they charge.
Now, I'm not a Cypress cheerleader, but I'm familiar with their EZ-USB ICs, and I think they're a good fit for this application.
Yes, they're surface mount (possibly socket-able) parts.
Yes, they need to be programmed with custom firmware.
And, yes, it requires somebody with Linux coding experience to write a Pi-side driver.
So, all of that puts a dent in the ability of users to implement DIY peripheral boards.
But what you get for all that work might make the tradeoff worthwhile.
(Cypress datasheet: http://www.cypress.com/?rID=38801
Up to 40 GPIO pins. An additional I2C bus controller. Two additional USARTs.
And, just by putting the Cypress chip in the loop, the RaspberryPi's Broadcom chip is effectively buffered, even if you don't put buffers on the I/Os of the Cypress chip (which you should).
All the additional GPIO and serial interfaces mean that you can implement some of the wishlist features like interfaces to external ADC and DAC chips, without sacrificing additional GPIOs on the RPi board itself.
And the issues surrounding bidirectionality of the RPi's GPIOs can be gotten around simply:
Either the I/O board driver can set the directionality of the Cypress I/O pins appropriately,
or you can actually split the Cypress GPIOs a group of dedicated inputs and a group of dedicated outputs, and have the driver send read and write requests to the appropriate Cypress pin.
That latter approach is highly "pin-intensive." But it permits extremely simple board construction, with unidirectional buffers on all the Cypress IOs, since their direction will now be fixed.
In terms of cost: it looks like the Cypress CY7C68013A chip will cost ~$6 from Digikey in qty~1000.
But, in terms of parts count, the board can (potentially) be as simple as one Cypress chip and an I/O header. The low-power Cypress chip can draw power from the +5V USB supply on the RPi board. If there is a layer of buffers on the Cypress I/Os, they would probably do well to get their own, independent, power supply.
Anyway. That's an alternative approach to multiplexing for the forum to consider.
We still need a solid design for a basic bidirectional buffer board to make the existing GPIOs usable, while protecting them from the slings and arrows of ESD, overvoltage, and short circuit events.