I am thinking about doing a quadruple whammy Arduino board: Atmel Mega 640 + CPLD + wireless + MicroSD card.
For those of you who are unfamiliar with the term "CPLD", it is a reconfigurable hardware chip. Essentially, today's CPUs all follow what is called the "von Neumann architecture" in which a CPU executes a set of instructions in a sequential fashion. A CPLD is very different. It is a "sea of logic gates", so it essentially executes every instruction simultaneously. So if you visualize a plot with the horizontal axis being what can be done per unit time, and the vertical being how much time to do a task, the CPU plot looks long and thin. It can't DO much per unit time, but you can string together many instructions to do complex tasks. The CPLD plot looks short and fat. It can DO a lot per unit time, but cannot do complex tasks since tasks must be completed in essentially 1 time unit. CPUs are much more well known both because they are programmed in a manner more intuitive to people and because ultimately you can get a lot more done with a CPU. However, their linear nature means that there are things that CPUs do not do well, like simultaneously and monitoring rapidly fluctuating inputs.
The reasoning behind integrating a wireless network and MicroSD is much simpler. First of all, the MicroSD card has a very simple interface (8 wires) so its easy to plop down and very useful for data recording, etc. Its so simple there is no reason to NOT integrate it. Purchased separately, a wireless network is expensive. You must buy a shield ($20-30) and a wireless module (another $20-$30). So this ends up being an additional $40-$60, not that expensive until you consider that the most interesting uses are to have not just one or 2 devices but to have a "cloud" of them. So for 10 devices, that would cost $500 -- outside of the casual hobbyiest budget. However the basic wireless chips cost about $4 for single quantities, and maybe $5 when the few additional parts are added. So an integrated wireless would cost an extra $50.
It seems to me that once you leave thru-hole chipsets, the casual user can't just plop the AVR down on his own handwired circuit board... but if people are OK with that (and it seems like they are given how many other surface mount Arduino clones exist), then it really opens the door for a base platform that hits key features that the Arduino is missing today. If the chips are integrated into the base platform, this platform becomes MUCH cheaper and smaller then buying all the shields separately. The final result is a tiny computer with the all the essential components of a "standard" PC (CPU, network, disk) AND the components required for a high performance/high IO embedded device (CPLD running at 100mhz with 50+ digital IOs). This combination can be used for sensor networks, cooperative mini-robotics, illumination, art/design projects etc and would probably cost around $50 or less.