Miracles on the servers.

Please pay attention to the fact that I consider this situation out of the ordinary, bordering on mysticism, and therefore I ask you to forgive me for the excessive literary character of the description of the problem.

Introduction.

I have a question. What do you think, if the program does not use a random number generator anywhere (In general, the application does not accept any incoming arguments that may affect its operation.), Can it, depending on the phase of the moon or something else, produce different results? Well, as if probably not, I think you agree with me.
Now let's imagine the situation.

We write:

#include <stdio.h>
void main ()
{
printf ("Hello, world!\n");
}

Without expecting any trick, we write gcc -o hello hello.c
Run
[~] $ ./hello
[~] $ Hello world

Joyful, you went on to go about your business. Three days have passed. You saw the binary, and decide to run it in order to enjoy your programmer’s genius once more.
')
[~] $ ./hello
[~] $ Hi man, that you have not been seen for a long time. What, wrote "hello world" and calmed down? Okay, pah at you, go ahead and go about your business, and I will go to read rsdn.

Yeah, submitted. There are two options, or stop drinking, or call 03. That's how it is.

Attention, the question.

Actually available:
Hosting provider, which turns 4 sites of friends and acquaintances.
Engine, lovingly written in Kohana.
Another site that is running on the next version of the engine.
So, this site periodically gives this error.
Picture
And the actual text:

system/libraries/drivers/Database/Mysql.php [61]:
mysql_connect() [function.mysql-connect]: Access denied for user 'dbuser'@'localhost' (using password: YES)

Neither the engine nor the server (according to the assurances of support) no one touches. No finger at all. On average, such an error can be observed 2-4 times a day with a total duration of 1-2 hours.
UPD: Naturally, the first thing that was done was that the mysql server was tested for performance, and it works and does not cough. Checked both from this server and remote, everything is fine. Well, that is, that it works, it’s understandable, they checked this user. You never know ...
And now, actually, why all this was written. I would like to hear any, even the most insane assumptions with what it can be connected.
UPD2: I summarize. The problem is either in our engine or in the server. The first is impossible because: 1) If the problem is in the engine, it may or may not work. Periodically, he can not work. Because for the last 4 days nobody touched the engine out of harm's way -> the problem is not in the engine. 2) This engine is spinning on 4 work projects and on 4 actively tested. Everywhere everything is fine. -> The problem is not in the engine.
The second one seems to be impossible, because 1) the support says that he did not touch the server. 2) All tests that are run when the engine is “down” tell us that everything is fine with the server.