Are UTF16 (as used by for example wide-winapi functions) characters always 2 byte long?
Short answer: No.
The size of a wchar_t—the basic character unit—is not defined by the C++ Standard (see section 3.9.1 paragraph 5). In practice, on Windows platforms it is two bytes long, and on Linux/Mac platforms it is four bytes long.
In addition, the characters are stored in an endian-specific format. On Windows this usually means little-endian, but it’s also valid for a wchar_t to contain big-endian data.
Furthermore, even though each wchar_t is two (or four) bytes long, an individual glyph (roughly, a character) could require multiple wchar_ts, and there may be more than one way to represent it.
A common example is the character é (LATIN SMALL LETTER E WITH ACUTE), code point 0x00E9. This can also be represented as “decomposed” code point sequence 0x0065 0x0301 (which is LATIN SMALL LETTER E followed by COMBINING ACUTE ACCENT). Both are valid; see the Wikipedia article on Unicode equivalence for more information.
Simply, you need to know or pick the encoding that you will be using. If dealing with Windows APIs, an easy choice is to assume everything is little-endian UTF-16 stored in 2-byte wchar_ts.
On Linux/Mac UTF-8 (with chars) is more common and APIs usually take UTF-8. wchar_t is seen to be wasteful because it uses 4 bytes per character.
For cross-platform programming, therefore, you may wish to work with UTF-8 internally and convert to UTF-16 on-the-fly when calling Windows APIs. Windows provides the MultiByteToWideChar and WideCharToMultiByte functions to do this, and you can also find wrappers that simplify using these functions, such as the ATL and MFC String Conversion Macros.
Update
The question has been updated to ask what Windows APIs mean when they ask for the “number of characters” in a string.
If the API says “size of the string in characters” they are referring to the number of wchar_ts (or the number of chars if you are compiling in non-Unicode mode for some reason). In that specific case you can ignore the fact that a Unicode character may take more than one wchar_t. Those APIs are just looking to fill a buffer and need to know how much room they have.
You seem to have several misconception.
There is a static type in C++, WCHAR, which is 2 bytes long. (always 2 bytes long obvisouly)
This is wrong. Assuming you refer to the c++ type wchar_t - It is not always 2 bytes long, 4 bytes is also a common value, and there's no restriction that it can be only those two values. If you don't refer to that, it isn't in C++ but is some platform-specific type.
There are no "extra wide" functions or characters types widely used in C++ or windows, so I would assume that UTF16 is all that is ever needed.
UTF16 seems to be a bigger version of UTF8, and UTF8 characters can be of different lengths.
UTF-8 and UTF-16 are different encodings for the same character set, so UTF-16 is not "bigger". Technically, the scheme used in UTF-8 could encode more characters than the scheme used in UTF-16, but as UTF-8 and UTF-16 they encode the same set.
Don't use the term "character" lightly when it comes to unicode. A codeunit in UTF-16 is 2 bytes wide, a codepoint is represented by 1 or 2 codeunits. What humans usually understand as "characters" is different and can be composed of one or more codepoints, and if you as a programmer confuse codepoints with characters bad things can happen like http://ideone.com/qV2il
Windows' WCHAR is 16 bits (2 bytes) long.
A Unicode codepoint may be represented by one or two of these WCHAR – 16 or 32 bits (2 or 4 bytes).
wcslen returns number of WCHAR units in a wide string, while wcslen_l returns the number of (locale-dependent) codepoints. Obviously, wcslen <= wcslen_l.
A Unicode character may consist of multiple combining codepoints.