For very high bandwidth real-time capture, PCI/PCIe is a good way to go.

For almost any other situation, it is way better to use USB, FW, eSATA or similar and implement the capture device as an external hardware dongle.

Note that the large bandwidth of gbit ethernet, eSATA or similar may allow even most high-bandwidth real-time capture products to be implemented as external hardware.

Remember that the market is moving away from stationary computers, so not only is the PCI an ageing technology on the way out, but more and more users will be without a machine for PCIe too.

Looking at laptops, PCMCIA and CardBus are old technologies. And people with older laptops do not have any Express Card slots. And whatever of these technologies you look at, the amount of power available (and ability to cool off the heat) is limited. The small edge size (unless a part of the card extends outside the laptop) means expensive contacts - especially if you need coaxial contacts for high-frequency of low-level signals.

A PCI or PCIe card may be problematic to build in case you need more significant amounts of shielding - remember that the inside of the PC is a quite hostile environment. And the different laptop cards are so small that you can't fit whatever filter components you may need.

With an external dongle with a hotplug-compatible interface allows you to decide on the physical dimensions. As already mentioned, you can select the most optimal power source based on the exact requirements - including battery operation for off-line or off-site data capture. An external dongle can make use of a 64GB flash card or a small (1.8" or smaller) HDD for quite extensive off-line data captures.

You can place the dongle close to the measurement point. You can add LEDs, displays or other forms of visual feedback on the box - try crawling on the floor to try to look at a LED on a PCI card - remember that gamers may have a window on their computer case, but a professional user doesn't, so you have a very small area on the back of the computer for connectors and any required LEDs.

Once upon a time, it was very easy/nice/cheap to build PC boards for the original PC bus or for the ISA bus introduced by AT-class machines (286 and forward). Many of us have bought PC prototype boards, where the PCB has buffer chips and some address decoding logic and then a large prototyping area for own designs. Today, the huge bandwidth, bus-mastering etc does increase the burden of developing an internal PC board. To that comes the requirements of writing drives. But drivers for 32-bit Wintel? Or 64-bit? Or for Linux? And the driver may have to think about cache coherency between multiple processors, or multiple (potentially hyper-threaded) processor cores. A large number of USB-connected devices can be designed as one of several classes of standard hardware, reducing the need for custom drivers. Your device may look like a high-speed serial port, or a disk drive, directly supported by newer operating systems.

In the end, you should really have strong reasons before continuing on the route of developing internal boards for a PC or a laptop. You should probably work with a very, very specific product that allows you to charge very high prices for each sold board. Or you should work with a high-volume mass-market product.