SD cards have very complex controllers and they are evolving in a manner that make them difficult to use with the Arduino. Higher end cards often perform poorly on the Arduino. Here is a bit of background. First, SPI mode is not used in most devices so the SPI controller is not very good. The flash erase groups in SD cards are very large, 128 KB is common. This means that rewriting file structures can result in a huge amount of data being moved. This is what is happening when you see long clock activity. The card indicates busy by holding data out low. The spec allow a card to go busy for up to 250 ms. Cards have two write modes, single block random mode and multiple block sequential mode. For file writes I must use single block mode. This is always slow but often extremely slow in high end cards. microSD cards also have poor support for this mode. I have several applications that use multiple block mode and they run much faster. The binaryLogger.pde example can log 40,000 16-bit samples per second without dropping data on a good SanDisk video card. Looks like I should develop a Serial logger using this method. It requires allocating a huge contiguous file and writing it using the multiple block sequential mode. This makes the app more complex and not so easy to understand. The write time for a 512 byte block in sequential mode is about 850 microseconds with occasional busy times of 2 - 3 ms on a SanDisk Extreme 30 MB/sec card. In random mode this card is often slower than five year old class 2 cards.