The initial and final permutation have no influence on security (they are unkeyed and can be undone by anybody). The usual explanation is that they make implementation easier in some contexts, namely a hardware circuit which receives data over a 8-bit bus: it can accumulate the bits into eight shift registers, which is more efficient (in terms of circuit area) than a single 64-bit register. This process "naturally" performs the initial permutation of DES.
In more details: Suppose that you are designing a hardware circuit which should do some encryption with DES, and receives data by blocks of 8 bits. This means that there are 8 "lines", each yielding one bit at each clock. A common device for accumulating data is a shift register: the input line plugs into a one-bit register, which itself plugs into another, which plugs into a third register, and so on. At each clock, each register receives the contents from the previous, and the first register accepts the new bit. Hence, the contents are "shifted".
With an 8-bit bus, you would need 8 shift registers, each receiving 8 bits for an input block. The first register will receive bits 1, 9, 17, 25, 33, 41, 49 and 57. The second register receives bits 2, 10, 18,... and so on. After eight clocks, you have received the complete 64-bit block and it is time to proceed with the DES algorithm itself.
If there was no initial permutation, then the first step of the first round would extract the "left half" (32 bits) which, at that point, would consist of the leftest 4 bits of each of the 8 shift registers. The "right half" would also get bits from the 8 shift registers. If you think of it as wires from the shift registers to the units which use the bits, then you end up with a bunch of wires which heavily cross each other. Crossing is doable but requires some circuit area, which is the expensive resource in hardware designs.
However, if you consider that the wires must extract the input bits and permute them as per the DES specification, you will find out that there is no crossing anymore. In other words, the accumulation of bits into the shift registers inherently performs a permutation of the bits, which is exactly the initial permutation of DES. By defining that initial permutation, the DES standard says: "well, now that you have accumulated the bits in eight shift registers, just use them in that order, that's fine".
The same thing is done again at the end of the algorithm.
Remember that DES was designed at a time when 8-bit bus where the top of the technology, and one thousand transistors were an awfully expensive amount of logic